The race for physical AI is shifting part of the technological pressure away from GPUs and data centers toward far less visible components. The permanent neodymium-iron-boron (NdFeB) magnets used in motors and actuators are one of them. If humanoid robots reach the manufacturing scale some companies are anticipating, securing neodymium, praseodymium, dysprosium, and terbium could become an industrial problem comparable to the supply of advanced chips.
Key facts about magnets for robots and AI in 20 seconds
- Humanoid robots need numerous compact, high-power motors, many of them based on NdFeB magnets.
- Adamas Intelligence expects robotics to become the leading consumer of these magnets by around 2040.
- China dominates much of rare earth processing and global magnet manufacturing.
- The United States is funding mines, separation, metallurgy, and new factories to reduce that dependence.
- Alternative motors and materials that could reduce future rare earth consumption are also advancing.
Until now, much of AI infrastructure could be explained through a fairly familiar chain: GPUs, HBM memory, high-speed networking, servers, data centers, and electricity. The arrival of robots, autonomous vehicles, and other machines capable of acting on the physical world adds a new layer.
A GPU can decide how a robotic arm should move, but it’s ultimately motors and actuators that have to carry out that movement. That’s where magnets come in.
From AI Model to Motor: The Hardware That Makes a Robot Move
NdFeB magnets combine neodymium, iron, and boron, and stand out for delivering a very high magnetic field within a relatively small volume.
That property explains their presence in electric motors, wind turbines, electronics, industrial automation, and defense systems. It also fits particularly well with robotics.
A humanoid needs to fit numerous motors into a limited space. Arms, shoulders, hips, knees, hands, and other joints all require actuators capable of generating force and executing precise movements without pushing the overall weight too high.
Neodymium and praseodymium provide the main magnetic properties. In applications that run at high temperatures, heavy rare earths such as dysprosium and terbium can be added to improve the magnet’s behavior under heat.
The problem appears when moving this architecture from a few thousand prototypes to millions of robots.
Adamas Intelligence believes robotics could become one of the biggest drivers of NdFeB magnet demand by 2040. The firm has even modeled what would happen under extreme humanoid-adoption scenarios.
One of them starts from the 10 billion robots Elon Musk has publicly floated as a long-term possibility. Adamas calculated that manufacturing that many by 2040 would require a volume of magnets equivalent to 186 times current global annual NdFeB production.
That figure is useful for gauging the scale of the challenge, but it shouldn’t be mistaken for a production forecast. Whether 10 billion humanoids actually get built is far from certain, and motor technology could also change well before reaching that kind of scale.
China’s Advantage Goes Far Beyond Its Mines
Talking about rare earths tends to bring mining straight to mind, but for the tech industry the problem runs deeper.
Between pulling ore out of the ground and installing a magnet in a robot lie several stages: concentration, chemical separation, oxide production, conversion into metals and alloys, and finally, magnet manufacturing and machining.
China has spent decades building industrial capacity across much of that chain.
This is especially sensitive for heavy rare earths. Dysprosium and terbium are used in certain magnet formulations intended for applications that need to withstand high temperatures while keeping their properties.
That geographic concentration introduces a risk similar to the one the tech industry already knows well from semiconductors: having access to the raw material doesn’t mean having the capacity to turn it into a usable component.
The United States is now trying to rebuild those intermediate steps.
The line Donald Trump used on July 15, when he publicly called on U.S. industry to manufacture magnets, simplifies a much broader industrial strategy.
Washington is backing projects spanning everything from extraction to finished magnet factories.
MP Materials Shows How the U.S. Wants to Rebuild the Chain
One of the most advanced projects belongs to MP Materials, owner of Mountain Pass in California.
The company is working to integrate U.S. extraction with magnet processing and manufacturing in Texas. Its relationship with the U.S. Department of Defense shows just how far this has moved beyond being treated as a purely mining business.
The agreement reached in 2025 sets a minimum price of $110 per kilogram of NdPr for ten years, under the terms established in the contract.
The U.S. government is also backing an increase in manufacturing capacity.
The mechanism tries to solve a historic problem for Western rare earth projects: spending years building facilities only to end up competing against Asian material sold at far lower prices.
Guaranteeing a price floor reduces some of that economic risk and gives more visibility to investments that need long periods to pay off.
The U.S. strategy also extends to dysprosium and terbium.
Serra Verde, a Brazilian producer that USA Rare Earth has agreed to acquire, has set minimum prices for its first-phase production: $575 per kilogram of dysprosium and $2,050 for terbium, alongside $110 for neodymium and praseodymium.
USA Rare Earth announced an agreement in April to buy Serra Verde for $2.8 billion. Since the deal has yet to close, the Brazilian company shouldn’t be considered fully integrated into the group just yet.
Elsewhere in the U.S., companies such as Noveon Magnetics have also raised financing specifically to expand domestic magnet manufacturing and reduce reliance on Chinese supply.
This string of deals shows something important for the tech industry: Washington isn’t just trying to create another mine. It wants a chain capable of ultimately producing motors and components.
AI’s Next Problem May Be Outside the Data Center
Robotics is substantially expanding the list of physical resources tied to artificial intelligence.
A data center needs semiconductors, memory, copper, fiber optics, transformers, cooling, and large amounts of electricity. A robot adds batteries, motors, gearboxes, sensors, bearings, and magnetic materials.
That’s why scaling physical AI presents different challenges than deploying a model in the cloud.
Building another million robots means repeating almost its entire bill of materials a million times over. Software can be copied at a very low marginal cost; an electric motor can’t.
Companies such as Tesla, Figure, Agility Robotics, Apptronik, and other humanoid makers are working precisely on cutting costs and designing robots that can be manufactured at industrial scale.
NVIDIA occupies a different position in this chain. Its platforms provide the compute, models, and tools for developing robots, but the growth of the market it’s targeting will also depend on manufacturers being able to secure every physical component they need.
That’s an important difference from purely digital generative AI.
GPU availability can be increased by building new semiconductor fabs. Supplying materials for millions of machines simultaneously requires mining, chemical plants, metallurgy, and industrial capacity spread across numerous suppliers.
Europe Is Also Trying to Rebuild Industrial Capacity
The response isn’t limited to the United States.
Europe has projects aimed at recovering part of its magnet processing and manufacturing capacity. Solvay, for instance, keeps rare earth separation activity running in La Rochelle, France, while Neo Performance Materials is building European magnet manufacturing capacity in Estonia.
Australia also holds a significant position thanks to its resources and companies such as Lynas Rare Earths, one of the leading producers outside China.
For makers of robots, vehicles, and industrial systems, this diversification has another advantage: it reduces dependence on a single country for a component that’s hard to replace quickly.
But building alternative capacity takes years. A chemical separation plant doesn’t simply appear because neodymium prices rise, just as a new semiconductor fab doesn’t start turning out advanced chips a few months after being announced.
Rare-Earth-Free Motors Could Change the Equation
There’s also a technological variable that rules out simply projecting today’s demand linearly out to 2040.
Not every electric motor needs rare earth permanent magnets.
Induction motors and electrically excited synchronous motors both exist, alongside research into new magnetic materials. Companies such as Niron Magnetics, for example, are working on magnets based on iron nitride.
The auto industry itself has already used different architectures specifically to cut costs or reduce dependence on certain materials.
Something similar will likely happen in robotics. If the price or availability of NdFeB becomes a serious constraint, manufacturers will have a strong incentive to redesign motors and actuators.
Recycling will also grow.
Vehicle motors, hard drives, industrial equipment, and other devices contain magnetic materials that can be recovered. Building a circular supply chain would make it possible to obtain some rare earths without going back through initial mining.
None of these alternatives eliminates the short-term importance of NdFeB magnets, but they do call for treating ten- or fifteen-year forecasts with caution.
Physical AI Forces a Look Beyond the GPU
The race for magnets shows just how much the development of artificial intelligence is starting to intertwine with industries that once seemed far removed from software.
During the first big wave of generative AI, the most sought-after component was the GPU. Then came the constraints around HBM memory, power capacity, transformers, and data centers.
Robotics introduces an even longer chain.
A humanoid needs compute to understand its surroundings, but also hundreds or thousands of physical components capable of turning its decisions into movement. Inside some of the motors that make that movement possible are a few grams of materials whose production remains heavily concentrated geographically.
The United States, Europe, Australia, and other countries are putting capital toward building alternatives. China starts with decades of industrial experience and capacity that can’t be replicated overnight.
The next battle in AI, then, won’t be fought only over who gets the fastest GPUs or trains the best model. A much less visible part of it will depend on who can manufacture, at scale, the motors, magnets, and materials that let that intelligence leave the data center and go to work inside a physical machine.
Frequently Asked Questions
Why do humanoid robots need rare earth magnets?
Many motors and actuators use NdFeB magnets because they deliver high magnetic density in a small space. That makes it possible to build compact, powerful systems suited to a robot’s many joints.
Could a magnet shortage limit robot manufacturing?
It could become a constraint if humanoid production grows much faster than the capacity to manufacture magnets and process rare earths. However, the industry is also developing alternative motors and materials.
Why is China so important in magnet manufacturing?
Its advantage doesn’t come only from mining. China has built extensive processing, separation, metallurgy, and magnet manufacturing capacity over several decades.
Are there robot motors that don’t use rare earths?
Yes. Different motor architectures without rare earth permanent magnets already exist, and new magnetic materials are also being researched. Adoption will depend on factors such as performance, size, cost, temperature, and industrial availability.
